241 results about "Function computation" patented technology

The embodiment of the invention provides a cross-chip platform instruction conversion method, which comprises the following steps: collecting instruction set information of a source chip platform and a target chip platform, establishing an instruction feature analysis model, and calculating a semantic similarity matrix between instructions in a feature vector space by adopting a kernel function; training a multilayer graph attention neural network based on the semantic similarity matrix, mapping feature vectors of a source platform instruction and a target platform instruction to a shared embedding space, and performing iterative optimization on a mapping relation graph by applying a reinforcement learning method to obtain an instruction equivalent conversion rule; the instruction equivalent conversion rule is loaded in the instruction conversion engine, the input source platform instruction is decoded and analyzed, and a converted target platform instruction sequence is output to an instruction execution unit of a target chip platform; the resource overhead in the conversion process can be optimized, and the overall efficiency of the system is improved on the basis of ensuring the conversion correctness.

The invention discloses a method and a system for establishing a comprehensive load model containing a distributed power supply, and relates to the technical field of load model establishment, after the establishment system obtains power system information through an API interface, a topological graph model of a power system is established, each component in the power system is abstracted as a node and an edge in the topological graph model, and the topological graph model is established; after simulation software is used for conducting simulation detection on the topological graph model, a target value of the topological graph model is calculated through an objective function, an optimal path between nodes is updated through a path algorithm, current distribution of edges is adjusted through a graph cut algorithm, power loss is reduced, overload is avoided, the steps are repeated for multiple iterations, and when convergence conditions are met, the optimal path is obtained. And outputting the final topological graph model as a comprehensive load model. The system can optimize the targets at the same time through a graph optimization algorithm, improves the overall benefits of the system, and overcomes the defect that the complexity of the power system cannot be comprehensively considered in single target optimization.

The invention provides an edge-end large language model reasoning acceleration method and device based on core particle and near memory calculation, and the method comprises the steps: storing a model weight in Flash, and completing the operation of all matrix vector multiplication in the whole reasoning process through the cooperation of the Flash and NPU. And in the calculation part in the Flash, the NPU needs to send the input vector to the Flash, the weight matrix stored in the Flash and the input vector are used for matrix vector multiplication, and a result is obtained and sent back to the NPU. And the calculation part in the NPU is as follows: an input vector is stored in the NPU, a weight matrix needs to be read block by block from the Flash, and the operation is completed in the NPU. And the operation result is stored in the NPU. The attention operation is independently completed by the NPU. The special function calculation is independently completed by the NPU. And the NPU and the Flash respectively bear the task proportions, so that the NPU and the Flash can better collaboratively complete the reasoning task of the large language model.

The invention belongs to the field of machine learning, and discloses an FPGA (Field Programmable Gate Array) superposition processor acceleration system and method based on state space duality, which comprises a sparse predefined data acquirer, a reconfigurable systolic array, a partial sum cache, an element-by-element operation cache, a function calculation module, an on-chip memory management module and the like. Zero elements are eliminated through the sparse predefined data acquirer, redundant data are reduced, and redundant calculation is remarkably reduced; the reconfigurable systolic array flexibly supports multiple calculation modes, and the utilization rate of hardware resources is increased; the part and the cache realize cross-cycle accumulation and element-by-element operation cache integration results, the function calculation module completes nonlinear operation, and the on-chip memory management module optimizes result cache, so that on-chip operation of SSD calculation is ensured, off-chip memory access is remarkably reduced, and reasoning efficiency and energy efficiency are improved; by adopting the system, the memory occupation is effectively reduced, the element-by-element calculation efficiency is improved, and the sparse calculation redundancy is reduced, so that the reasoning process of the Mamba2 model is remarkably accelerated.

The invention relates to the technical field of numerical control programming, in particular to a five-axis machining path planning method and system based on data driving, and the method comprises the following steps: obtaining real-time coordinates of each axis of a machine tool, calculating linear velocity and angular velocity components to construct a Jacobian matrix, executing singular value decomposition, and calculating a conditional number ratio by using maximum and minimum singular values; and inputting a nonlinear mapping function to calculate a dynamic penalty factor, generating a rotating shaft weighted item in combination with a rotating shaft identifier, constructing a weighted damping least square objective function, calculating a five-axis motion increment, and accumulating the five-axis motion increment with a real-time coordinate to generate a target absolute position coordinate. According to the method, the pose singularity degree is quantified by monitoring the machine tool pose condition number ratio and converted into the dynamic penalty factor to apply the self-adaptive constraint to the rotating shaft, the severe sudden change of the rotating shaft in the singularity area is inhibited, the tool nose track following error is minimized, and meanwhile smooth distribution of the motion increment is achieved; and the dynamic stability and the surface quality of five-axis linkage machining are improved.

The invention belongs to the technical field of data visualization, and particularly relates to a dynamic visualization generation method, system and device based on user behaviors and context awareness and a medium, and the method comprises the steps: generating a structured operation sequence after standardizing original input data; performing time sequence modeling on the operation sequence to generate a user behavior semantic vector, and generating a joint feature vector by adopting an attention mechanism; pre-loading an input data set, and generating a candidate view set in combination with the user intention preference in the joint feature vector; calculating view scores through a layout optimization function, and screening an optimal view scheme with the highest score; dynamically simplifying or enhancing the chart component according to the equipment performance and context parameters in the joint feature vector; calling a rendering engine to output a view, and monitoring a user interaction event; and updating the strategy network through a PPO algorithm, and optimizing a subsequent view generation path. The problem of unsmooth interaction caused by a static template is avoided, and the user experience is remarkably improved.

The invention discloses a program fault processing method, device and equipment of a software system and a storage medium, relates to the technical field of computers, can be applied to financial science and technology and medical health business scenarios, and comprises the steps that a fault knowledge graph of the software system is acquired, the fault knowledge graph is dynamically constructed based on multi-source heterogeneous data, and the fault knowledge graph is stored in a database; wherein a diversified atlas relationship is stored; performing graph neural network driven association analysis based on the system component dependency relationship and the first association relationship, mining a hidden fault propagation path on a component dependency link, and determining a fault association component and a fault root cause corresponding to the program fault data based on the hidden fault propagation path; based on the fault association component, the fault root cause and the second association relationship, determining a plurality of candidate repair schemes corresponding to the program fault data; and calculating evaluation scores of the plurality of candidate repair schemes through the utility function, sorting according to the evaluation scores, and determining an optimal repair scheme of the software system about the program fault in the plurality of candidate repair schemes.

The invention discloses an LLC resonance parameter forward design method, device and equipment, a medium and a program product, and the method comprises the steps: constructing a resonance mathematical model according to the information of elements participating in resonance in an LLC resonant converter, constructing a phase plane model of the LLC resonant converter based on the resonance mathematical model, constructing a target function according to the phase plane model, and carrying out the forward design of the LLC resonance parameter. Calculating state trajectory parameters based on the objective function, and performing parameter design on the LLC resonant converter according to the state trajectory parameters; according to the LLC resonant converter, the phase plane model of the LLC resonant converter is constructed, so that the state track simulation of the LLC resonant cavity is realized, the design of the LLC resonant parameters is realized through the coordinates of the state points on the phase plane, the parameter design complexity of the converter is effectively reduced, the parameter design efficiency is improved, and the performance of the LLC resonant converter is effectively improved.

The invention provides a multi-task model fusion method and device, a medium and a program product. The method comprises the following steps: acquiring a plurality of expert models of different task scenes; taking the parameter of each expert model and the sparse version corresponding to each expert model as an initial particle swarm; calculating a performance score of the fusion model for each task according to a corresponding preset evaluation index, and taking a mean value of the performance scores of all the tasks as an objective function of particle swarm optimization; iteratively updating the initial particle swarm by adopting a particle swarm optimization algorithm, calculating a historical optimal solution and a global optimal solution of each particle based on a target function, and dynamically adjusting the particle speed and parameters; and taking the global optimal particle parameters after iteration as fused model parameters, generating a fusion model, and deploying the fusion model at a target terminal. According to the method, the model fusion efficiency and the multi-task processing capability are improved, and the occupation of the computer storage space is reduced.

The invention discloses an asynchronous federal learning method and system based on dynamic contribution perception and old adaptation, and the method comprises the steps: receiving a corresponding contribution degree after a server broadcasts a latest global model and a contribution degree vector; judging whether the received contribution degree meets a training condition, if so, downloading a latest global model for training, and if not, continuing to use an old version global model for training; fusing the loss function, the model deviation penalty and the old suppression item, and constructing a local optimization objective function; calculating a dynamic learning rate; and executing gradient descent to update the local model, and sending the local model to the server, so that the server performs weighted aggregation on the local model by using the dynamic aggregation weight after receiving the client local model greater than or equal to a preset number, updates the client contribution degree, and performs normalization to generate a contribution degree vector. The method gives consideration to the communication efficiency and the model precision, is suitable for a large-scale distributed machine learning scene, and has the advantages of high robustness and good adaptability.

The invention provides a hardware-friendly high-precision floating point transcendental function calculation system, and the system comprises a parameter obtaining module which is configured to obtain a to-be-calculated single-precision floating point parameter and a calculation function selection signal; the transcendental function preprocessing module is configured to perform preprocessing on the single-precision floating point parameters respectively and output a target preprocessing result; the calculation module is configured to select a target preprocessing result and perform calculation according to the target preprocessing result to obtain a target polynomial calculation result; the leading zero detection module is configured to perform leading zero detection on the target polynomial calculation result to obtain a detection result; and the transcendental function post-processing module is configured to perform standardization processing on the detection result and output a target result corresponding to the transcendental function. According to the system, the problems that an existing transcendental function calculation scheme is low in efficiency and long in calculation period are solved.

The invention discloses an opportunity fair federal learning excitation method and system based on the game theory, and the method comprises the steps: a model owner determines a reward value of a current round, determines the contribution rate of participating in the training of the current round according to each data node of the reward value of the current round, and carries out the local model training corresponding to each data node; the model owner aggregates the local models of the data nodes according to the reputation values of the data nodes to obtain a global model; constructing a first utility function according to the overhead of each data point in the current round, and constructing a second utility function according to the earnings of the model owner; and calculating the utility of a model owner and each data node according to the first utility function and the second utility function, and adjusting the contribution rate of the next round of training and the reward value of the next round according to the utility result. The optimal balance between the power data node retraining contribution rate and the training cost is realized, and the training effect and the duration of federal learning are ensured.

The invention relates to the technical field of graph neural networks, in particular to a graph neural network architecture optimization method based on grammar genetic programming, and the method comprises the steps: S1, constructing a grammar rule library of a graph neural network architecture GNN; s2, based on the grammar rule base, generating an initial GNN architecture population through genetic programming; s3, predicting the performance of the individual GNN architecture in the population by using a Gaussian process proxy model; s4, searching an optimal GNN architecture in the syntax tree space through a genetic algorithm; s5, for the individual GNN architecture in the population, calculating a training probability based on an expected improvement function of the individual GNN architecture, if the training probability is higher than a preset evaluation threshold, selecting the individual GNN architecture to carry out actual training evaluation, obtaining the authenticity performance of the individual GNN architecture, and updating the Gaussian process agent model by using the authenticity performance; and S6, judging whether a preset optimization termination condition is met or not, if so, outputting the optimized GNN architecture, and otherwise, returning to the step S3. According to the method, the automation degree, efficiency and model performance of GNN model architecture search can be improved.

The invention relates to a communication method, device and system. A computing node obtains a code of a first function based on a session establishment request of a terminal device, wherein the session establishment request is used for requesting to establish a first session. And after the code of the first function is acquired, the computing node receives a first request for accessing the first function from the terminal equipment through the first session, and operates the code of the first function according to the first indication information to process the first request. According to the embodiment of the invention, the computing node can obtain the code of the first function before receiving the first request from the terminal equipment, so that the code downloading process does not need to be executed after the first request is received, the time delay of function computing is greatly reduced, and the efficiency of function computing is improved.

The present invention relates to a VDF (initial state, t), comprising: a one-way invertible function S component directly or indirectly constructing the VDF, wherein the S component is defined as an invertible function where a computation delay of forward logic is significantly greater than that of reverse logic; a computation phase: a intermediate states, and / or F(A0, Ab)=S(A0⊕Ab, △) indirectly constructing the VDF; and a verification phase: receiving the initial state and a+1 states, preferably completing a+1 VDF-1 verification tasks by means of parallel (a+1)*(b?b:1) S-1 logic verifications. A VDF construction with salt is preferred; the S component preferably has a structure of F(A, B)=g(A⊕B), wherein g-1 uses sparse tap design. Disclosed are proposed industrial standards VDF32 and VDF8192. The present invention achieves the effects of no initialization parameter phase (Setup); quantum-resistant design; and entropy preservation. Disclosed related uses comprise a randomness beacon, a portal website, a proof-of-replication, a storage service terminal, a cloud computing terminal, two types of blockchain systems, Token encoding and verification, and automatic commitment opening. To support industrial implementation, further disclosed are a VDF computing apparatus, a VDF verification apparatus, a medium, and a chip.

The embodiment of the invention provides a nonlinear function calculation method and a special function calculation unit (SFU). The method comprises the following steps: determining a to-be-fitted function and a fitting interval corresponding to the to-be-fitted function; based on the to-be-fitted function and the fitting interval, configuring segment parameters and linear parameters corresponding to the to-be-fitted function; and performing fitting calculation based on the piecewise parameters and the linear parameters to obtain a first target function corresponding to the to-be-fitted function, and completing nonlinear function calculation based on the first target function. According to the invention, the problems of low precision, poor performance and low efficiency of SFU processing in the prior art are solved, and the effect of improving the SFU processing precision, efficiency and performance is achieved.

The invention discloses a cloud edge-end function calculation unloading method capable of cooperatively transmitting data. According to the method, a function calculation unloading process is divided into a data transmission process and a function calculation process. The function calculation process depends on data transmission among the nodes, so that a collaborative data collaboration mechanism can be introduced in the data transmission process, and the task calculation time delay is further reduced. Compared with the prior art, the method has the advantages that a data unloading mechanism is introduced in the cloud side-end collaborative computing unloading process under the complex function dependency relationship; task calculation time delay is reduced, and calculation efficiency is improved.

The invention discloses an intelligent stereoscopic warehouse inventory data synchronization method and device and a computer program product, and the method comprises the steps: collecting multi-source data of an intelligent stereoscopic warehouse, and constructing a multivariate data set; embedding the multivariate data set into a nonlinear mapping space, and constructing an embedded state vector for expressing inventory behaviors; a scheduling weight evaluation function is constructed based on the embedded state vector and historical error feedback, the scheduling weight of the inventory operation is calculated through the scheduling weight evaluation function, a decision is made on the synchronization mode of the inventory operation according to the scheduling weight, and the synchronization mode comprises instant synchronization, delay synchronization or merging synchronization; and based on the scheduling weight and historical error feedback, dynamically correcting the response tension coefficient of each dimension in the embedded state vector, and applying the corrected response tension coefficient to subsequent embedded state vector construction to form a feedback closed loop. According to the invention, a powerful guarantee is provided for efficient, accurate and stable operation of the intelligent warehousing system.

The invention belongs to the technical field of deep learning, and particularly discloses a model post-training quantification method and system based on Pareto optimization. Through the application, the weight and activation in the pre-trained floating point model are converted into integers with low bit width, the size and precision of the model are simultaneously optimized by utilizing the Pareto thought, and the comprehensive loss of multiple targets is minimum by constructing the joint loss function and calculating the Pareto plane, so that a Pareto optimal state is achieved; and then a balance point with the shortest actual reasoning time is screened out as a final quantification result, so that the improvement of the compression ratio of the model and the reasoning acceleration on resource-limited equipment are realized on the premise of ensuring the precision of the model.

The embodiment of the invention provides a data processing method, which comprises the following steps of: converting an e index calculation task into an index calculation task aiming at a target fixed point number, and splitting the index calculation task aiming at the target fixed point number into a shift task and index calculation aiming at a target decimal part, an index calculation task aiming at the target decimal part is approximate to linear function calculation (namely addition and multiplication based on calculation parameters), so that the calculation process can be greatly simplified, and simple multiplication and addition are executed by utilizing a digital signal processor based on the calculation parameters corresponding to the target decimal part; compared with the prior art, the method has the advantages that operation steps and required computing resources are reduced, all computing resources of one digital signal processor are fully utilized in the process, the whole task process can be performed in series, judgment branches and data rereading do not exist, the overall computing efficiency is improved, and hardware resource consumption is reduced.

The invention provides a high-performance parallel implementation method based on Transform transplantation of a SW processor, and relates to the technical field of processor calculation parallel implementation, slave core calculation is performed by using a SW Athread programming method, each independent function of forward propagation and backward propagation of a ViT algorithm is rewritten into a unified function, a calculation intermediate variable in the forward function is simplified, and the calculation efficiency is improved. Transplanting operation of the ViT algorithm is realized; according to the method, an intermediate variable in a forward function calculation process is simplified, a ViT algorithm based on a Transform architecture is successfully transplanted and operated on a domestic SW processor, unnecessary redundant calculation is removed through an optimization algorithm, and the operation efficiency of a ViT model on a Shenwei super computer is remarkably improved.

The invention provides a Serverless function calculation method and a Serverless function calculation system with instance level capacity expansion configuration. The method comprises the following steps: receiving a capacity expansion threshold value and a concurrency upper limit configuration parameter of a function instance; monitoring the number of concurrent requests of a single function instance in real time; when the number of concurrent requests reaches a capacity expansion threshold value, triggering an asynchronous capacity expansion operation; requests exceeding the concurrence upper limit are distributed in a balanced mode through multi-level request scheduling; executing batch capacity expansion and contraction operation based on the length of the message queue; and implementing asynchronous request processing. The system comprises a controller, a configuration module, a monitoring module, a capacity expansion module, a scheduling module, a batch capacity expansion and contraction module and an asynchronous request processing module. The method has the beneficial effects that through instance-level capacity expansion configuration and multi-level request scheduling, the resource utilization rate and the request processing efficiency of the Serverless function are remarkably improved, meanwhile, the cold start delay is reduced, and the flexibility and the expandability of the system are enhanced.

The invention discloses a quantum bit mapping method, device and equipment, a medium and a computer program product, and the method comprises the steps: calculating initial quantum bit mapping according to the mutual relation between quantum bits on which each logic gate in a quantum circuit depends; the initial quantum bit mapping is initial mapping from logic quantum bits to physical quantum bits; generating a to-be-executed gate circuit set according to the execution dependency relationship among the logic gates in the quantum circuit; based on the initial quantum bit mapping, the execution cost of each to-be-executed gate circuit in the to-be-executed gate circuit set is calculated, and the to-be-executed gate circuit with the minimum execution cost is selected as a target execution gate circuit; and based on a preset heuristic cost function, calculating costs of different SWAP strategies of the target execution gate circuit, and determining optimal quantum bit mapping according to the SWAP strategy with the minimum cost. According to the invention, the insertion of the SWAP gate in the quantum circuit is significantly reduced, so that the efficiency and precision of quantum bit mapping are improved.

The invention discloses a robot generative adversarial self-imitation learning method based on large language model feedback, and belongs to the technical field of computer systems based on specific calculation models, and the method comprises the steps: (1) initializing a strategy network, a discriminator network and a demonstration pool; (2) constructing and training a large language model feedback network: obtaining a training sample of the large language model feedback network, and training the large language model feedback network according to the training sample; and (3) an iterative learning step (33): updating the strategy network according to the reward calculated by the new cost function. According to the robot generative adversarial self-imitation learning method based on large language model feedback, the advantages of the large language model in the aspects of semantic comprehension, reasoning and knowledge organization are used as a potential award priori source, and a stable award signal can be provided when the demonstration data quality is poor, so that a strategy exceeding original demonstration is learned.

The invention provides a large language model knowledge fusion method and device based on parameter restriction, and relates to the technical field of knowledge fusion. The method comprises the steps of obtaining a plurality of task models; performing redundant parameter suppression on the task vectors in the plurality of task models by adopting a nonlinear activation function to obtain the suppression degree of redundant parameters; calculating the similarity of the same position parameters in the initial task vector by adopting a nonlinear activation function; obtaining a parameter importance score in the task vector according to the suppression degree and similarity of the redundant parameters; according to the parameter importance score in the task vector, obtaining a final combined task vector; and the finally combined task vectors are zoomed again, and combined task model parameters are generated. By adopting the method, the big language model merging performance can be improved.

The invention discloses a hardware-based SOFTMAX function calculation method and device. The hardware-based SOFTMAX function calculation method comprises the following steps: receiving a multi-dimensional input vector corresponding to an SOFTMAX function through an input port; uniformly dividing the multi-dimensional input vector into a plurality of data segments according to an element sequence through a channel distributor, and distributing the data segments to corresponding operation channels; performing an exponential operation on the data segment using the operation channel to generate a local exponential vector; aggregating the local exponent vectors with an accumulator to generate a global exponent sum; distributing the global index sum to each operation channel, so that the operation channels generate output sub-vectors according to the local index vectors and the global index sum; and splicing the output sub-vectors according to the element sequence through a recombination unit to generate an SOFTMAX output vector. The efficient SOFTMAX function hardware acceleration calculation under the multi-channel parallel pipeline processing architecture is realized, and the calculation performance and throughput are improved by several times compared with the traditional serial software implementation.

The invention discloses a background vehicle anthropomorphic trajectory generation method for an automatic driving test scene, and the method comprises the following steps: extracting a plurality of groups of driving features in an interaction scene according to a set interaction logic, and carrying out the simulation training, so as to construct a driving simulator, driving characteristics in each group of interaction scenes are used as driving behavior data of a plurality of background vehicles, set driving styles are configured in a one-to-one correspondence manner, and a corresponding driving strategy is set for each driving style so as to construct a dynamic Bayesian game framework; the method comprises the following steps: acquiring a driving style of a background vehicle in any group of interaction scenes in real time to carry out iterative training on a multi-target utility function, calculating a profit value corresponding to the driving style of the background vehicle by utilizing the multi-target utility function, and selecting a corresponding driving strategy from a dynamic Bayesian game framework to generate a track control instruction; and performing personification optimization on the trajectory control instruction by using a random disturbance mechanism, thereby realizing multi-dimensional driving behavior dynamic generation and complex interaction logic accurate modeling.

The invention discloses a function calculation mechanism for document generation, which belongs to the technical field of complex logic content generation and comprises the following steps of: introducing a JavaScript engine: utilizing the JavaScript engine as a core processing unit for function calculation, providing strong script execution capability and wide library support by the JavaScript engine, and generating a function by using the JavaScript engine; various complex logic calculation and data processing can be efficiently executed; a Word template engine is used, a POI-TL is used as a template engine for document generation, the POI-TL is based on an Apache POI library, and a Word document can be conveniently read and written in; according to the method, efficient calculation is carried out, namely, the JavaScript engine can quickly execute complex logical operation and data processing, so that the document generation time is shortened, for example, all calculation and document generation tasks can be completed within several seconds when a contract containing complex calculation and logical judgment is generated; and automatic processing: high automation is realized in the whole process from data input to document generation, manual intervention is reduced, and the overall working efficiency is improved.

A method, computer program product, and computer system of artificial intelligence (AI) intervention to detect and mitigate an abnormality associated with an object moving during performance of a process. A trained recurrent neural network (RNN) determines, from sensor data, a probability (Pr1) that the abnormality existed at a first time, where Pr1 exceeds a threshold T1 and in response, an alternative generative adversarial network (AGAN) determines a probability (Pa1) that the abnormality existed at the first time. A score S1, which is computed as a function of Pr1 and Pa1, exceeds T1 and in response the RNN and the AGAN determines, from the senso data, a probability (Pr2) and a probability (Pa2), respectively, that the abnormality existed at a second time. A score S2, which is computed as a function of at least Pr2, Pa2, and (S1-T1), exceeds a threshold (T2) and in response, the abnormality is mitigated.

The invention relates to the technical field of artificial intelligence and privacy protection, and discloses a vertical field large model privacy forgetting method and terminal based on double Fisher matrix guidance, and the method comprises the steps: separating two text sequence data subsets from an original training corpus of a vertical field large language model, namely a reserved data set and a forgetting data set; the reserved data set contains vertical field professional data which does not contain privacy information, and the forgotten data set contains privacy sensitive data to be eliminated. Based on an information geometry theory, calculating a reserved Fisher information matrix and a forgotten Fisher information matrix of model parameters on the reserved data set and the forgotten data set respectively; based on collaborative analysis of a reserved Fisher information matrix and a forgotten Fisher information matrix, a parameter forgotten guide function is constructed, guide function values of all parameters of a model are calculated, and a standardized guide vector is constructed according to the guide function values. According to the method, refined balance of privacy forgetting of the large language model in two levels of privacy elimination and capability keeping is realized.